The present study compares human translation of Arabic taboo expressions found in Woman at Point Zero with translations of the same expressions generated by ChatGPT and DeepSeek. The dataset included 20 taboo expressions that were manually identified in the text of the novel. First, a quantitative analysis was carried out to evaluate the quality of machine translations of the expressions in question according to the metrics of the Multidimensional Quality Metrics (MQM) framework. The analysis encompassed such dimensions as accuracy, fluency, terminology, and style. In addition, eight selected examples that represent the major categories of taboo language, as well as the most common difficulties in translation of the latter, were examined qualitatively. As a result, the present study provides one of the very few systematic assessments of the ability of generative AI to translate taboo expressions from Arabic literature. Human translation turns out to be superior to machine translation in terms of maintaining semantic accuracy, cultural significance, and pragmatics of taboo language. While ChatGPT and DeepSeek produce fluently written and terminologically accurate translations, they tend to neutralize, weaken, or misinterpret culturally encoded messages, thus reducing the stylistic and emotional effect of the source text. It appears that contemporary generative AI technologies are unable to cope with the translation of literary language that is culturally sensitive and requires interpretation in a particular context. One of the limitations of the study is its qualitative analysis based on a number of selected examples taken from one particular literary text. Therefore, the findings cannot be generalized to any other literary genre or language pair. However, the research yields original results for the rapidly growing field of AI-assisted literary translation, as it involves comparative MQM assessment of human and large language model translations of Arabic taboo expressions and emphasizes the role of human post-editing in literary translation.
The present study compares human translation of Arabic taboo expressions found in Woman at Point Zero with translations of the same expressions generated by ChatGPT and DeepSeek. The dataset included 20 taboo expressions that were manually identified in the text of the novel. First, a quantitative analysis was carried out to evaluate the quality of machine translations of the expressions in question according to the metrics of the Multidimensional Quality Metrics (MQM) framework. The analysis encompassed such dimensions as accuracy, fluency, terminology, and style. In addition, eight selected examples that represent the major categories of taboo language, as well as the most common difficulties in translation of the latter, were examined qualitatively. As a result, the present study provides one of the very few systematic assessments of the ability of generative AI to translate taboo expressions from Arabic literature. Human translation turns out to be superior to machine translation in terms of maintaining semantic accuracy, cultural significance, and pragmatics of taboo language. While ChatGPT and DeepSeek produce fluently written and terminologically accurate translations, they tend to neutralize, weaken, or misinterpret culturally encoded messages, thus reducing the stylistic and emotional effect of the source text. It appears that contemporary generative AI technologies are unable to cope with the translation of literary language that is culturally sensitive and requires interpretation in a particular context. One of the limitations of the study is its qualitative analysis based on a number of selected examples taken from one particular literary text. Therefore, the findings cannot be generalized to any other literary genre or language pair. However, the research yields original results for the rapidly growing field of AI-assisted literary translation, as it involves comparative MQM assessment of human and large language model translations of Arabic taboo expressions and emphasizes the role of human post-editing in literary translation.